stitch-design-taste
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided project intent and vibe descriptions to synthesize a
DESIGN.mdfile, which serves as a set of instructions for a downstream AI agent. This functions as a multi-step prompt chain. - Ingestion points: User-provided descriptions of project intent and design atmosphere in
SKILL.md. - Boundary markers: No explicit delimiters are present to isolate user-derived content from the generated system instructions.
- Capability inventory: The skill is limited to generating text-based design specifications and lacks capabilities for command execution or network exfiltration.
- Sanitization: No explicit input validation or escaping mechanisms are described.
- [EXTERNAL_DOWNLOADS]: The skill references legitimate external services for its primary function.
- Evidence: Mentions Google Stitch (
labs.google/stitch) and thepicsum.photosservice for placeholder image generation.
Audit Metadata